Summary:

BF-43773 tracks an 11.6% performance regression in Atlas Infinite vs Atlas Core for aggregation workloads. WT-14895 is called out in the BF as a potential fix. This ticket builds a proof of concept to measure the performance impact of WT-14895's layered cursor optimizations before committing to a full implementation.

Background:

Layered cursors split inserts into two steps (search, then insert), whereas classic cursors combine them. This split adds a small API overhead (~0.41%). Reverting to the combined classic behavior is non-trivial, so we need to confirm the net performance benefit before investing in a full fix.

Scope:

• Implement a minimal, non-production PoC.
• Measure performance impact on the BF-43773 workload.
• Estimate how much of the 11.6% regression is recovered.
• Document complexity, risks, and cost to productionize.